cremind llm

cremind llm is the CLI for managing the LLM side of Cremind: which providers are configured, what models each one exposes, which models the agent reaches for at high and low reasoning effort, and the GitHub Copilot device-code OAuth flow. It is the terminal counterpart to the Settings → LLM Providers page in the web UI.

CREMIND_TOKEN is required for every subcommand, and all of them accept the root-level --json flag. Provider secrets such as API keys are stored server-side and never read back — they surface only as a configured: yes flag.

Subcommand groups

The command splits into three sets:

Providers

providers list prints every provider the server knows about, with whether it is configured, how many models were discovered, and the active auth method:

cremind llm providers list

NAME       DISPLAY              CONFIGURED  MODELS  ACTIVE_AUTH
anthropic  Anthropic            yes         8       anthropic
openai     OpenAI               no          0
copilot    GitHub Copilot       yes         12      oauth_personal

providers configure sets a provider's configuration — typically the API key and the active auth method, plus any extra fields via --json. At least one of --api-key, --auth-method, or --json must be supplied:

# Standard API key
cremind llm providers configure anthropic --api-key sk-ant-...

# Switch the active auth method without touching the key
cremind llm providers configure copilot --auth-method oauth_business

# Extra fields (e.g. base URL, organization) via JSON
cremind llm providers configure openai --json '{"organization":"org-...","base_url":"https://api.openai.com/v1"}'

providers models <provider> lists the model ids a provider exposes, and providers delete-config <provider> removes every stored config field for it (the provider stays known but its configured flag flips to no).

Model groups

Cremind picks from two model groups: a high group for heavyweight reasoning and a low group for lighter work such as the skill classifier. Read them with get:

cremind llm model-groups get

{
  "default_provider": "anthropic",
  "model_groups": {
    "high": "anthropic/claude-opus-4-7",
    "low":  "anthropic/claude-haiku-4-5-20251001"
  }
}

Update any subset with set — omitted fields keep their current value:

cremind llm model-groups set --high anthropic/claude-opus-4-7
cremind llm model-groups set --default-provider openai --high openai/gpt-5

Device-code auth

device-code start begins a device-code OAuth flow (currently for GitHub Copilot) and prints a verification URL, a short user code, and an opaque device_code:

cremind llm device-code start

verification_uri  https://github.com/login/device
user_code         ABCD-1234
device_code       4fe...e8c
expires_in        900
interval          5

Open the URL, enter the code, then block on device-code poll until the browser flow completes. The token is stored server-side:

cremind llm device-code poll 4fe...e8c

complete (token stored server-side)

poll blocks while the upstream returns pending (Ctrl-C is safe and does not invalidate the code), and exits with an error if the code expires — run device-code start again to get a fresh one.

Related